Classifying Gaze Patterns for Preemptive Screening of Autism Spectrum Disorder (ASD) in Children Using Deep Learning

Autism Spectrum Disorder (ASD) is characterized as a neurodevelopmental condition that persists throughout life. The incidence of autism in children has continued to rise manifold, emerging as a daunting challenge to existing public health systems. The onset of Autism is prenatal, and discernible atypical behavioural patterns emerge in the first 12 – 18 months of life. Despite the growing awareness about autism, the absence of definitive biomarkers impedes the early diagnosis of autism in children. While research in early diagnosis of autism continues to evolve, there is some academic consensus that gaze patterns in children with autism vary significantly compared to their neurotypical peers. It is in this context that our research explores the viability of using gaze patterns of children as a biomarker for autism, using the “Saliency4ASD” dataset. Our study demonstrates that neural network architectures outperform traditional machine learning algorithms. Early diagnosis can facilitate early interventions resulting in better outcomes for children on the autism spectrum.